Information Technology Reference
In-Depth Information
Acquiring novel algorithms represents an endeavor to plista. In contrast to Netflix,
plista's item collections are subject to continuous changes (see Sect.
6.4
). Thus, an
algorithmwhich performs well on news of two months ago could provide inadequate
suggestions today. We cannot guarantee that an algorithm will achieve similar per-
formance on novel items. As a result, plista created a platform providing researchers
and practitioners with access to actual interactions. The platform was first released
in 2010 as the “Open Recommendation Platform” (ORP) for internal usage. ORP
allowed plista's recommendation engineers to conveniently add novel recommenda-
tion algorithms to their eco-system. Three years later, plista opened the platform for
interested researchers and other third parties to evaluate their recommendation algo-
rithms. Moreover, the platform supported plista to stay connected with the research
community and actively exchange ideas. ORP ought to provide a representative
selection of news portals. Otherwise, evaluations may include biases toward certain
aspects. Thus, plista directly included two large-scale general news portals along
with a selection of minor, rather topic-specific clients. ORP enables participants to
interact with real users in a real-time setting. Interaction takes place in a two-stage
process. First, news portals visitors load a news page initiating a request for recom-
mendations. Second, the participants' server receives the requests and returns a list
of suggested news items. The news portal embeds the list in the news page shown
to the visitor. This setup reflects a genuine use-case. Methodologically, we refer to
such settings as “living labs”. This is due to the unpredictability of future interac-
tions. Note that ORP represents a subset of all news portals served by plista. Having
the idea of ORP in mind, plista contacted publishers with whom they had long-term
relationships. Insightful discussion covered both advantages and disadvantages of
data sharing with and contributions by researchers. Plista managed to include a rep-
resentative group of publishers into ORP. The group of publishers comprises minor,
medium, and large scale news portals. Furthermore, the news topics cover general
selections as well as news portals providing news for specific subjects. The selection
contains some news portals which operate on a similar regional level allowing evalu-
ating recommendation methods which exchange information between domains. The
included publishers use different types of widgets. Thus, ORP allows us to eliminate
biases due to graphical user interfaces to a certain degree. These biases include posi-
tion relative to the news article and the number of recommendations among others.
We describe major components as well as vital aspects of ORP in the following
subsections.
6.3.1 Involved Parties
News recommender systems concern different interest groups. These groups include
news portal operators, content providers, advertising companies, recommendation
providers, and visitors amongst others. We outline the individual perspective of each
group.
Search WWH ::
Custom Search